FPGAs offer many advantages and opportunities for use as accelerators. This project initially started as an approach to accelerate computer vision algorithms, in particular OpenVX graphs, in collaboration and sponsored by Intel. We have also started investigating the use of FPGA accleration to support graphs processing and neurobiological applications.

The goal of this project is to extend and improve the performance of OpenVX kernels and to investigate how an OpenVX graph (aka a graphics pipeline) can be partitioned between the host and the FPGA and accelerated. It has several thrusts:

Identify best approaches to accelerate individual kernels using either HLS or OpenCL and compare quality of resulitng "code", programmer productivity, and power consumption.

Develop several non-trivial pipelines using OpenVX and investigate how to best partition/map these to FPGAs

Compare performance and power consumption vis a vis GPUs.

This project is supported by Intel Corp.